UNIVERSITY OF APPLIED SCIENCES
OFFENBURG
HOCHSCHULE FÜR TECHNIK UND WIRTSCHAFT
Theory and Applications
of
Digital Image Processing
A. Erhardt  Ferron
A demo version of the complete course can be found at
http://www.dip-seminar-online.com/english/
1
1 The Image Processing System
An image processing system (fig. 1.1) consists of a light source to illuminate the scene,
a sensor system (usually a CCD -camera) and an interface between the sensor system
and the computer. Among other things, the interface converts analog information into
digital data which the computer can understand. This takes place in a special piece of
illumination camera
output
device
frame
grabber
host
1.1: Components of an image processing system
hardware, the frame grabber, which also stores the image. Many types of frame grabber
hardware are supplied with special signal processors, so that very calculation-intensive
parts of the image processing programs can be run in a time-efficient way. Usually the
frame grabber package contains a library of often-used routines which can be linked to
the user s program. The results of an image processing run will be transferred to the
outside world by one or more I/O interfaces, the screen and the normal output devices
like printer, disks etc.
The classical configuration of image processing hardware is not a stand-alone system
but has to be directed by a host computer. However, the newest developments are able
to integrate the complete image processing system into the camera.
In this module we will talk about the hardware components of image processing sys-
tems. You will receive the basics which will enable you to conceptualize an image
processing system along with the knowledge necessary to be able to compare the capa-
bility and the compatibility of components offered by different companies You should
be familiar with the terminology in the field of personal computers. In addition, some
Copyright 2000 A. Erhardt-Ferron
2 1 The Image Processing System
knowledge of algebra is required for part 1.2.3 of this unit.
1.1 Illuminating the Scene
An important aspect of image processing is the proper choice of light source, which
has to be appropriate to the system s working environment. A good choice of illu-
mination will allow the image processing system to receive the best image under the
circumstances and the number of procedures necessary for image restoration will be
minimized. The goal is to optimize the dynamics and the contrast of an image. This
means that an object has to be photographed with a maximum number of intensity steps
and should, at the same time, have the best possible contrast with its background.
By the choice of the light source the features of the radiation (e.g. wavelength, di-
rection of oscillation, spatial intensity distribution), can be selected depending on the
requirements of the object s surface (i.e. structure, color, transparency etc.). In any
case, however, the aim is to establish a homogenous and temporally constant illumina-
tion over the whole area of interest.
Daylight is usually not very well suited to illuminating a scene for image processing
because the color and the intensity of the light depend on the time of day, the time
of year and the weather conditions. Similarly ill-suited is the uncontrollable light
in a production line in a factory hall. Situations where uncontrolled light cannot
be avoided, for example in the environment of autonomous moving vehicles, will
always provide challenges for the image processing system.
Tungsten light sources are very cheap but not very well suited for image processing,
especially if the image readout frequency of the camera is not a multiple of the net
frequency (i.e. 50 Hz or 60 Hz). This is often the case with cameras with features
other than those required by the video norm, e.g. line cameras. In this case,
light frequency and the image readout frequency differ, resulting in undesired
interference, which appears as lighter and darker stripes on the screen and which
reduces the image quality. Of course, they could be operated with direct current
but an additional drawback is a non-uniform illumination field along with the fact
that they get very hot.
Fluorescent lamps have a large homogenous illumination field. They can be oper-
ated with frequency rectifiers to prevent a modulation of the light intensity and
the resulting interference. In addition, they do not get very hot. One possible
disadvantage could be the spectral limitation which is provided by the fill, but
depending on the application this might even be a desirable feature. Indeed, fluo-
rescent lamps are therefore often used to illuminate scenes for image processing.
Quartz Tungsten Halogen (QTH) lamps do not have the problem with mains fre-
quencies. Like normal tungsten lamps they have a tungsten filament inside,
which starts to glow when connected to electricity. However, opposed to normal
Copyright 2000 A. Erhardt-Ferron
1.1 Illuminating the Scene 3
tungsten lamps, they are filled with a rare gas and a small amount of a halogen,
(mostly iodine or bromine compounds). When the lamp is turned on the follow-
ing thermo- chemical process called halogen cycle takes place:
The tungsten atoms, which are emitted from the hot filament (3300 C)
cool down at some distance to a temperature below 1400 C. There they
chemically react with the halogen - atoms. This chemical compound is
gaseous down to a temperature of 250 C.
With the thermal current of the halogen gas the compound molecules get
close to the hot tungsten filament, where they are divided up into their parts
- tungsten and the halogen.
The tungsten attaches to the filament, the halogen is free for a new repetition
of the process.
Because of this eternal rejuvenation of the filament, the temperature of the fila-
ment can be much higher than the one of a normal tungsten lamp, and the inten-
sity does not vary much during one period of the alternating current. The result is
that halogen lamps are light sources with almost constant light intensity. They are
usually not directly applied to the scene, but mainly used as feeding light sources
for fiber optical systems.
Using fiber optical systems to illuminate small objects makes it possible to adjust
the angular distribution of the light intensity exactly to the requirements of the
task. In addition, areas which are hard to access can be illuminated properly. The
disadvantages with fiberoptics are the fact that about 40% of the light intensity is
lost by scattering and reflection effects and the relatively high price of the lamps.
Discharge lamps have very high radiation densities, a temporally constant luminosity
and their electromagnetic spectrum shows continuous or discrete lines, depending
on the illuminating gas. Certain kinds (flashlights) can be used for stroboscopic
illumination. However, they are also relatively expensive.
Light Emitting Diodes(LED s) react instantly and almost without inertia to contol
light intensity over a very wide range. This also makes them suitable for stro-
boscopic applications. Another advantage is their good monochromatic nature
which particularly suits them to situations where the chromatic aberration of the
camera objective plays a role. Additionally, they are reasonably priced, inex-
pensive to operate, and they are small and lightweight. Their lifetime of about
100 000 hours makes them practically maintenence-free. Furthermore, because
the use of LED s is not accompanied by heat, noise, vibration or high voltage,
their application range in industrial image processing has expanded dramatically
in recent years. Diodes are often arranged in arrays or as ring lights. In addi-
tion, like halogen lamps, they are used as feeding light sources for fiber optical
systems.
The emission of monchromatic light which is an advantage of LED s can, in some
situations, be disadvantageous.
Lasers have a high radiation power focused on a very small area; laser light is highly
chromatic and coherent. Nowadays, though, because of safety considerations
Copyright 2000 A. Erhardt-Ferron
4 1 The Image Processing System
(among other reasons) large laser units which scan scenes have been replaced by
laser diode modules. A laser diode module is the end result of a laser diode, elec-
tronics and optics built into a common housing. A laser diode module is about
the size of a thumbnail and, like an LED, can be integrated into systems with
limited space to be used as laser source. With laser diode modules it is possible
to project lines, points, circles, matrices of points, etc. Therefore the mechanical
adjustment of the object before recording the image can also be supported opti-
cally. Like halogen lamps and LEDs, lasers can be used as feeding light sources
for fiber optical systems.
Infrared light sources are always used in scenes where it is impossible to eliminate
unwanted influence of surrounding daylight or of scattered radiation from other
light sources. If, in addition, an infrared camera is used with a daylight blocking
filter, the influence of the ambiguous ambient light can be completely eliminated.
1.2: Construction of a ring light
If constant light intensity over a very long time period is required, the aging process
in all light sources has to be taken into consideration. The aging process causes a
decrease of intensity, and in most cases, the frequency spectum will shift towards longer
wavelengths.
If fibre-optic light sources are used to illuminate small objects, it is possible to direct
the angular distribution of the light flow explicitly and to fit the spatial distribution of
the illumination strength according to the object. In addition, difficult-to-access areas
can be illuminated. Fig. 1.2 and fig. 1.3 show some realizations. Light sources for the
coupling of fibre-optic units include, among others, halogen lamps, discharge lamps,
LED s and laser diodes. Because of scattering and reflection phenomena on the inner
boundaries of the optical fibres, the intensity losses amount to about 40%.
Fast-moving objects have to be illuminated stroboscopically. The synchronization
will be handled by the frame grabber hardware. There, the trigger signal for the camera
as well as for the stroboscope has to be created.
Depending on the positions of the camera and the light source, one distinguishes
among four fundamental ways of lighting a scene: incident light illumination, transmit-
Copyright 2000 A. Erhardt-Ferron
1.1 Illuminating the Scene 5
1.3: Various fiberoptic illuminations for image processing like point lights, ring lights, area lights
ted light illumination, light-field, and dark-field illumination.
Incident light illumination: Camera and light source are on the same side of the ob-
ject. The image shows the distribution of the light intensity reflected by the ob-
ject.
Transmitted light illumination: Camera and light source are on opposite sides of the
object. The screen shows the dark form of the object in front of a light back-
ground. Transmitted light illumination is applied when an object can be described
by its own form.
Light-field illumination: As with incident light, the camera and the light source are
positioned on the same side of the object. The part of the light which is directly
reflected into the camera will be used for the imaging process. Light-field illumi-
nation will show dark objects against a light background.
Dark-field illumination: As above, camera and light source are on the same side of
the object, but only the scattered light is captured by the camera. Dark-field
illumination produces a dark background with light objects.
We intuitively use light- and dark-field illumination to see a scratch on a record or a
CD. If it is held against the light, the scratch can appear dark on a light background
(light-field illumination) or light on a dark background (dark-field illumination).
If the four basic types of illumination are used with additional mechanisms, numerous
other possibilities to light a scene are available.
Copyright 2000 A. Erhardt-Ferron
6 1 The Image Processing System
camera light source camera
object
object
light source
incident light illumination transmitted light illumination
camera camera
light source
light source
object object
light field illumination dark field illumination
1.4: Basic illumination setups
light source
object
line mask
camera
structured light
a b
1.5: Structued illumination to measure three-dimensional objects[7]
a) illumination setup
b) projected lines
Copyright 2000 A. Erhardt-Ferron
1.2 Imaging Methods and Sensor Systems 7
Diffuse Lighting: If the surface of an object to be illuminated reflects strongly, direct
lighting cannot be applied. Diffuse light can be used instead, as might be gener-
ated by an overcast sky. The direct light is focused on a diffusing surface which
can be something as simple as a white sheet. The result is that only the scattered
light hits the object. Diffuse lighting  softens light conditions of a scene and
prevents strong reflexes.
Structured Illumination: This is applied when a three-dimensional object has to be
surveyed in two dimensions. Lines or a grid are projected onto the three-dimensional
form. The curvatures of the projected lines on the three-dimensional surface de-
pend on the position of the camera, the light source and the grid, and, of course,
the three-dimensional form of the object, which can be calculated from the bend-
ing of the lines.
Using the shadow of an object: If an object has about the same lightness as its back-
ground, it is often impossible to see it in the picture. If the object is three-
dimensional, the light source should be placed in a position allowing the shadow
of the object to appear in the background. Instead of the object itself, the shadow
will be measured. From the relative positions of camera and light source, the real
dimensions of the image can be calculated.
1.2 Imaging Methods and Sensor Systems
The term  image processing suggests that the pictures which will be processed are taken
by camera. This is often the case, but generally, every sensor which produces spatially-
distributed intensity values of electromagnetic radiation which can be digitized and
stored in RAM is suited to image capturing.
Various image capturing systems are used, depending on the application field. They
differ in the
acquisition principle
acquisition speed
spatial resolution
sensor system
spectral range
dynamic range
Apart from the area of consumer electronics, most apparatuses are very costly. The
greater the need for accuracy, the more hard- and software is used in the image cap-
turing system. The following list shows the most- used units for capturing images
electronically:
area scan cameras
Copyright 2000 A. Erhardt-Ferron
8 1 The Image Processing System
line scanners
laser scanners
computer und nuclear magnetic resonance (NMR) tomographs
thermographic sensor systems (e.g. infrared cameras)
ultrasonic devices
CCD sensors play a central role in most image processing systems. They are part of a
complex system which makes it possible to take images in problematic environments
with the necessary quality and accuracy.
Sensors can be categorized into the following classes according to their sensitivity
ranges:
Electromagnetic sensors for
gamma radiation
X-ray radiation
the visual spectrum
the infrared spectrum
the radio wave range
Each electromagnetic sensor is only sensitive to a certain range of electromagnetic ra-
diation. Other sensors like
ultrasonic sensors
magnetic sensors
may also be used for imaging, but they do not work according to the CCD principle.
However, in the context of this course, only the most important acquisition methods
will be considered.
1.2.1 CCD Cameras
In a film camera the photo-sensitive film is moved in front of the lens, exposed to light,
and then mechanically transported to be stored in a film roll.
A CCD camera has no mechanical parts. The incoming light falls on a CCD (Charge
Coupled Device) sensor, which consists of numerous light-sensitive semi-conductor
elements, the so-called pixels. They are arranged in a line (line camera) or a matrix
(area scan camera).
The image sensor is the heart of a digital camera. High resolution and color accuracy
as well as the signal-to-noise ratio depend on the quality of the CCD sensor. The physics
of a CCD sensor is the inner photo effect. This means that the incoming photons pro-
duce electrons in the semi-conductor material, which are separated in the photo diode
and stored as in a capacitor. This capacitor is connected to the surrounding electrical
Copyright 2000 A. Erhardt-Ferron
1.2 Imaging Methods and Sensor Systems 9
circuit via a MOS transistor, which acts like a light switch. If it is open, the charges will
be collected in the capacitor ( integrated) and will be transported when the switch is
closed. The number of electrons which are collected is proportional to the light which
reaches the light-sensitive part of the sensor. The exact physical processes, however,
are not of much relevance to our topic; you will be referred to the numerous literature
sources in opto-electronics, for example [5] or [2].
The Video Norm
Real time systems are usually based on video norms, which means that the image ac-
quisition as well as the conversion of the digital data into a video signal has to conform
to international standards. In Europe, this norm is defined by the Comité Consultatif In-
ternational des Radiocommunications (CCIR); in the USA, the norm is called RS-170
StandardindexRS170and was defined by the Electronics Industries Association (EIA).
The PAL(Phase Alternation Line)and SECAM (Sequentiel Couleur á Memoire color
standards are based on CCIR while the color system based on RS-170 is NTSC (Na-
tional Television System Committee). The fundamentals of all video standards reach
back to the times of tube cameras and tube monitors and seem a little archaic in the age
of CCD chips and LCD screens.
Both norms require the so-called interlace process, for the image on the screen to be
non-flickering (fig. 1.6). This means that a complete image (frame) is separated into
two half images (fields). One field consists of the odd lines; the other consists of the
even lines of the image. The electronic ray starts in the upper lefthand corner. After it
reaches the end of the first line (this takes 52 s (CCIR)), it goes back to the beginning
of the third line which takes 12 s (CCIR). During this time the horizontal sync signal
(H-Sync) is added to the video signal, which starts the new line. The front and back
porch of the line blanking pulse is used as a reference of the color black. In this way
first field second field
1
2
3
4
5
6
.
.
.
.
.
.
1.6: Interlace process: two fields make a frame
the electronic ray scans the first field with the odd-numbered lines. Then the vertical
sync signal (V-Sync) is added, which indicates the beginning of the next field. The V-
Sync - Signal is a more complex signal, which requires 50 video lines. Subsequently,
the second field with the even numbered lines is scanned. A complete scan with two
Copyright 2000 A. Erhardt-Ferron
10 1 The Image Processing System
fields consists of 625 lines and takes 40 ms. Because the initialization of the next field
takes 50 of the 625 lines, only 575 per frame or 287.5 lines per field are visible. In
Table 1.1: The video norms CCIR and EIA
CCIR RS-170
frame setup interlace interlace
color system PAL/SECAM NTSC
elds per sec 50 60
time per eld 20 ms 16.6 ms
time per frame 40 ms 33.3 ms
total number of lines 625 525
time per line 40 ms/625 = 64 s 33.3 ms/525 = 63.5 s
line frequency 1/64 s = 15.625 kHz 1/63.5 s = 15.750 kHz
information per line 52 s 52.7 s
line sync 12 s 10.8 s
eld sync 3.25 ms 50 Zeilen 2.54 ms 40 Zeilen
no. of visible lines 575 485
image format (horizontal:vertical) 4:3 4:3
pixel per line 575*4/3 = 767 485*4/3=647
time per pixel 52 s/767 = 67.8 ns 52.7 s/674 = 78.2 ns
pixel frequency 1/67.8 ns = 14.75 MHz 1/78.2 ns = 12.8 MHz
no. of line pairs 767/2 = 383.5 647/2 = 323.5
horizontal resolution 14.75 MHz/2 = 7.375 MHz 12.8 MHz/2 = 6.15 MHz
(max. Video frequency)
channel width 5 MHz 4.2 MHz
both video norms the ratio of width to height is 4:3. This means that there are 767 pixels
per line. This number of pixels is scanned in 52 s and results in a pixel frequency of
14.75 MHz.
However, for our visual perception a pixel has little relevance. Therefore video res-
olution is traditionally defined differently. The video system under the CCIR Norm
reaches its physical limitation when it has to show a test pattern of 383.5 black and
white line pairs, since two neighboring pixels then have to show the lowest and highest
intensity value. 383.5 periodic intervals in 52 s result in a frequency of 7.375 MHz,
which is the maximum possible bandwidth of any video component. On the other hand,
the CCIR - Norm allows a much lower video bandwidth (channel width) of 5 MHz.
Thererfore, the number of vertical lines which a camera can capture is a quality mea-
sure of the resolution. This parameter is called TV-Lines. It counts the number of
vertical, not the horizontal lines. Table 1.1 shows the specifications and their values
according to the video norms CCIR and RS-170
Norming video electronic units has the advantage that components such as CCD chips
from different manufacturers can be integrated into appliances. Norming is one of the
reasons for the vast increase in the use of CCD cameras in the entertainment industry,
including the private sector. Accordingly, the components have become affordable for
a wide range of users.
On the other hand, norming can also be restrictive. For example, the interlace process
required by the video norm ensures a flicker- free image but there are applications where
it has some disadvantages. If a fast- moving object is filmed by a norm video camera,
the object will have moved during the first 20 ms which were required to record the first
field. Therefore the beginning of every second line is shifted. This is known as comb
Copyright 2000 A. Erhardt-Ferron
1.2 Imaging Methods and Sensor Systems 11
effect.
Another disadvantage of the video norm is the short integration time, which is max-
imum 20 ms. If light conditions are unfavorable, it is impossible to produce pictures
of acceptable quality, even when all the amplification possibilities are applied. In a
case like this you use a camera with a long-term integration capability, which does not
conform with the video standard.
For applications where the video norm s disadvantages are too great there are norm-
free cameras. Progressive xcan cameras are an example which do not apply the interlace
process but instead scan the lines subsequently. These sensors are also able to use the
full resolution of modern graphic cards. Typical image formats are, for example, the
VGA resolution with 800 600 pixel or the SuperVGA format with an even higher
resolution of 1280 1024 ixel.
Cameras which do not conform to the video standard are usually more expensive
because they cannot be mass-produced.
High Definition Television (HDTV)
The original idea for the HDTV format (High Definition Television) came from wide
screen movies. In the early 1980 s Sony and NHK (Nippon Hoso Kyota) developed an
HDTV - film recording system (called NHK Hi- vision), which could be used to take
a scene and play and edit it immediately afterwards. This eliminated the many delays
which occur with normal film production. In addition, the new medium made a number
of special effects possible, which were impossible to do in traditional film production.
In addition, it turnded out that the movie audience were more impressed with wide
screen movies, because they create the impression that the viewer is part of the scene.
Soon efforts took place to develop a similar format for the television screen. The moti-
vation was less the enhancement of resolution but rather
the creation of a natural viewing experience by using the total human visual field
the absence of visible distortions like the comb - effect with the interlace method
high image quality.
Now, the developers of HDTV have had the same problems as the engineers involved in
the introduction of color television in 1954. Worldwide there are 600 million TV sets,
and the question arose, if HDTV should be compatible to the old standard, if it should
supplement the old standard or if the two standards should be transmitted simultane-
ously. The main problems were
the enormously high data rate of more than 40 Mbit/s, which requires a high
bandwidth or highly sophisticated compression techniques
the larger screen format, which meant that the users had to replace their old TV
screens with new and more expensive ones, if they wanted to benefit from the
new format
the fact that the new standard was incompatible with the PAL - System. This was
probably the biggest obstacle to the introduction of HDTV in all the the countries
which, like Germany, use PAL.
Copyright 2000 A. Erhardt-Ferron
12 1 The Image Processing System
the fact that the expensive equipment used by TV studios would have to be re-
placed by even more expensive equipment and, on the consumer side, peripheral
units like video recorders would also have to be replaced.
the increased image quality which would have to be assured at the production site
the marketing problem: unlike in Japan, users in Europe and in the United States
have to be  convinced that it makes sense to throw their old TV equipment out
and replace it with HDTV.
As a result, the introduction of analog HDTV has been quite difficult. It has had com-
pletely different receptions in Japan, the US, and Europe.
Japan has a quick start:
 1964 Fundamental research and development of HDTV starts
 1979 First TV broadcast in HDTV format
 1981 HDTV is officially introduced, which causes an  HDTV-shock in the
US and Europe.
 1989 Regular HDTV broadcasts in MUSE (the Japanese HDTV analog for-
mat) begin
 1997 Announcement to change to digital HDTV
Therefore, research, development and production of cameras, recorders, TV sets,
broadcasting systems etc. are much more advanced today than in the US or Eu-
rope. In addition, the Japanese have gained extensive knowledge and experience.
Japan is the only country in the world which broadcasts more than 9 hours per
day in the HDTV format.
The US counters:
 1977 Foundation of a study group concerning HDTV (SMPTE)
 1983 Foundation of the Advanced Television Systems Committee (ATSC)
 1986 The USA decides to support the Japanese system
 1989 The USA decides to drop the support for the Japanese system
 1990 Introduction of the digital HDTV-System DigiCipher
 1995 Agreement of the Grand Alliance on a common HDTV-Standard
 1997 Official start of HDTV broadcasting through terrestrial frequencies
with OFDM and 8-VSB
The introduction of HDTV in the USA was made difficult by the facts that too
many suggestions for systems existed and that the agreement on  the best system
took quite a while.
Europe is asleep:
 1986 Start of development of HD-MAC (the European HDTV analog for-
mat)
 1988 introduction of HD-MAC prototypes
Copyright 2000 A. Erhardt-Ferron
1.2 Imaging Methods and Sensor Systems 13
 1991 HD-MAC is dropped and a European Launching Group (ELG) is
founded to support the development of a European digital standard
 1993 The ELG is transformed to the Digital Video Broadcasting Group
(DVB). Norms for digital TV broadcasting, based on the MPEG2 compres-
sion method are developed.
 1994 Norms for satellite and cable transmittance are introduced
 1996 The norm for terrestrial transmission is introduced.
After dropping HD-MAC, analog HDTV was no longer supported in Europe.
Rather, a common broadcasting standard for digital TV was sought, which would
include digital HDTV based on MPEG2. However, the regular broadcast of
HDTV programs is not to be expected in the near future. Instead, several pro-
grams with a quality similar to that of the PAL system are supported.
The fields for the HDTV-format in Europe are mainly the areas of medical imaging,
the military, design, graphics, the print media, advertising, art and the movie industry.
Although TV is expected to support the HDTV format in the far future, experts estimate,
that not more than 20% of all programs will ever be broadcast in HDTV.
For industrial image processing this development has come too late. For areas which
are not suitable for a videonorm camera, numerous special developments have been
made, which are moderately priced for industrial standards.
Table 1.2: Comparison of HDTV - Norms of Japan, the USA und Europe
HDTV Japan HDTV USA HDTV Europe
frame setup interlace progressive scan progressive scan
number of lines 1125 1050 1250
visible number of lines 1080 960 1000
image format 16:9 16:9 16:9
(horizontal:vertical)
optimum viewing distance 3.3 picture height 2.5 picture height 2.4 picture height
vertical viewing angle 17 23 23
at optimum viewing distance
orizontal viewing anglel 30 41 41
at optimum viewing distance
CCD Sensor Architectures
CCD area scan cameras are available in several CCD sensor architectures, of which
three are currently on the market.
The term architecture refers to the way the information of the individual pixels is
bundled and integrated into a serial data stream. For all architectures there are camera
versions, which conform to the video norm and others, which define their parameters
freely. Descriptions of the above-mentioned architectures follow. However, not all
available cameras are included. In this sector there are many in- house developments
and developments for specific applications.
Copyright 2000 A. Erhardt-Ferron
14 1 The Image Processing System
The Interline Transfer Sensor: An interline transfer sensor is subdivided into light-
sensitive and storage areas (fig. 1.7). These are arranged in stripes. The charges
shaded
shift registers
(storage area)
light sensitive
sensor area
(illumination area)
output register
1.7: The Interline concept: illumination. and storage area are arranged in a stripe pattern
are integrated in the light-sensitive cells and subsequently transferred in a very
short time (about 2.5 s) to the shaded vertical shift registers. From there they are
transferred one line at a time to a horizontal readout register and then sequentially
to the input area of the video input unit (fig. 1.8).
With the interline transfer sensor the active light-sensitive area takes only a small
part of the total sensor cell. The connectors between the cells as well as the
shaded areas are not light-sensitive. This results in the fact that interline transfer
CCD cameras in the traditional architecture are much less light-sensitive than, for
example, frame transfer cameras, which are described below. There are several
developments which aim to offset this disadvantage. The lens -on - chip - technol-
ogy is worth mentioning. There, every sensor cell gets a micro lens, which bun-
dles the light normally falling on the shaded storage area into the light-sensitive
sensor area. This results in a sensitivity increase of about a factor of 2.
The Frame Transfer Sensor: On a frame transfer sensor the light-sensitive areas and
the storage area are located in two different blocks. The total area (light-sensitive
cells and shaded shift register) is about twice as big as on the interline transfer
sensor fig. 1.9). The total charge is shifted through the transport register into
the shaded shift register. From there it gets transferred into the horizontal output
registers and added to the serial data stream (fig. 1.10). Most frame transfer
CCD sensors also conform to the video norm.
But as with other architectures, there are forms which do not adhere to the video
norm.
Copyright 2000 A. Erhardt-Ferron
1.2 Imaging Methods and Sensor Systems 15
1st step 2nd step 3rd step
1.8: Transport of charges with the Interline Transfer CCD sensor
1st step: integrated charges are transferred to the vertical shaded readout registers.
2nd step: integrated charges are transferred to the horizontal readout register.
3rd step: serial output of charges
light sensitive
sensor area
(illumination area)
shaded
shift registers
(storage area)
output register
1.9: The Frame Transfer concept: illumination and storage area are two blocks
Copyright 2000 A. Erhardt-Ferron
16 1 The Image Processing System
1st step 2nd step 3rd step
1.10: Transport of charges at the Frame Transfer CCD sensor
1st step: integrated charges are transferred to readout registers.
2nd step: integrated charges are transferred to the horizontal readout register.
3rd step: serial output of charges
The Full Frame Transfer Sensor: ( The full frame transfer sensor, unlike the frame
transfer and the interline transfer sensor, does not incorporate a storage section.
The total sensor area is light sensitive (fig. 1.11). After the integration time the
the shutter is closed and the charges are read out line after line (fig. 1.12). Full
frame sensors always need a shutter camera. This sensor type has no influence
on the integration time. An external shutter takes care of that. The full frame
transfer image sensor makes fast data rates possible. It is mostly used in time-
critical applications. High resolution cameras (500 500 . . . 4000 4000 pixels)
also use a full frame transfer image sensor.
In this section, only the principal detector architectures have been described. There
are countless variations and combinations. Research is moving in the direction of the
development of so-called  intelligent cameras which will be capable of taking over
computer functions. Some projects of new camera developments will be discussed in
chapter 1.2.2. Research labs have been working on cameras with the ability to adapt to
light changes, cameras with stereoscopic vision, cameras with integrated capabilities of
smoothing and edge detection, etc. While most cameras which are currently available
are still based on the interline transfer technology[21][20], conform to the CCIR norm
and transfer images with the interlace mode, in the near future they will probably exceed
the ability of the human eye[12].
CCD Chip Formats
CCD chips are available in different formats (fig. 1.13). The designation of chip
sizes goes back to tube cameras. Typical outside diameters of these tubes are 1 inch,
Copyright 2000 A. Erhardt-Ferron
1.2 Imaging Methods and Sensor Systems 17
light sensitive
sensor area
(illumination area)
output register
1.11: The Full Frame Transfer concept: the total sensor area is light sensitive
1st step 2nd step
1.12: Transport of charges at the Full Frame Transfer CCD sensor
1st step: after the integration time the camera shutter is closed and the charges are
transferred to the horizontal readout register
2nd step: serial output of charges.
Copyright 2000 A. Erhardt-Ferron
18 1 The Image Processing System
1/4  CCD - Chip 1/3  CCD - Chip 1/2  CCD - Chip
8.0 mm 4.8 mm
6.0 mm 3.6 mm
4.0 mm 2.4 mm
3.2 mm
4.8 mm
6.4 mm
1  CCD - Chip
2/3  CCD - Chip
15.875 mm 9.525 mm
11.0 mm 6.6 mm
8.8 mm 12.7 mm
1.13: Chip sizes.
2/3 inch and 1/2 inch. A tube with a 1 inch outside diameter has a rectangular active
window measuring 16 mm diagonally. This format was retained for CCD sensors.
1 inch chips are very rarely applied nowadays; 1/3 inch and 1/2 inch chips, on the other
hand, are finding more and more applications, especially in the field of security cameras,
miniature cameras and home video cameras. In metrology however, the 2/3 inch chip
is predominant and will be in the foreseeable future.
Pixel sizes fall between 4 m 4 m and 16 m 16 m; the number of pixels range
between 500 500 in security cameras up to 5000 5000 in sophisticated measuring
applications[3].
Camera Configurations
There is a wide range of camera configurations, depending on their application field.
Camera types can differ in the way the pixels are arranged as well as in their spectral
sensitivity.
The part of the electromagnetic spectrum to which a camera is sensitive depends
on the semiconductor material the CCD chip is made of. As mentioned in previous
sections, the incoming photons produce free charge carriers by lifting the electrons of
the semiconducter material from the valence band to the conduction band. The number
of electrons produced is proportional to the number of incoming photons.
The spectrum of CCD sensors ranges from ultraviolet light up to the infrared area.
The spectral sensitivity depends on the energy gap E between the valence band and
the conduction band. For E = 1 eV, for example, an upper limiting wavelength of
= 1.24 m can be calculated from the equations
g
h > E and
c =
Copyright 2000 A. Erhardt-Ferron
1.2 Imaging Methods and Sensor Systems 19
h c
! < = max (1.1)
E
with:
E: energy difference between valence band and conduction band in eV
h =6:6262 10 34 Js: Planck s constant
: wavelength of light
: cut-off wavelength
max
: light frequency
c =299:8 106 m/s: speed of light in vacuum
1eV = 1:60219 10 19J: conversion factor
Table 1.3 shows various energy gaps and the resulting upper limiting wavelengths for
various semiconductor materials. From this table and from fig. 1.14 it can be concluded,
for example, that silicium is very well suited for the near infrared (IR-A) and the vi-
sual part of the electromagnetic spectrum, while for the far infrared spectrum (IR-C),
semiconducter materials with a lower gap between valence band and conduction band
should be used.
Table 1.3: Energy gap between valence and conduction band with the cut-off wavelength for
various semiconductor materials. Values at T=300 K. (Nr. 1 [8], Nr. 2 [13], Nr. 3 [14] Nr. 4 [9]
Nr. 5 [1] Nr. 6 [16] Nr. 8 [4] Nr. 9 [17])
Nr. Semi Conductor Chem. Abbr. E max
in eV in nm
1 indium - antimonide InSb 0.18 7754
2 lead telluride PbTe 0.311 5904
3 lead sulphate PbS 0.42 3351
4 germanium Ge 0.664 1879
5 silicium Si 1.1242 1107
6 gallium arsenide GaAs 1.424 867
7 cadmium - selenide CdSe 1.7 729
8 gallium phosphat GaP 2.272 553
9 cadmium sul de CdS 2.485 512
0,001nm 0,1nm 400nm 1mm
infrared
gamma-
gamma- X-ray UV- visible radio
radiation
IR-A IR-B IR-C
780nm 1400nm 3000nm 1mm
1.14: The electromagnetic spectrum
Copyright 2000 A. Erhardt-Ferron
20 1 The Image Processing System
CCD cameras for the visual spectrum are sensitive in a range of 400 nm up to 1000 nm
with a maximum of approximately 530 nm (green).
The following description shows cameras with various pixel arrangements.
Line cameras consist of only one line of CCD sensors. They are applied to problems
where a higher resolution is required or only one object dimension has to be sam-
pled. These days, line cameras with 8000 or more pixels and a pixel frequency
1.15: Various line cameras
with more than 30 MHz are series products. To minimize losses during the shift-
ing process of the electrical charges, line cameras read out towards both sides
of the line (fig. 1.16). This keeps the readout times low but also causes low
shaded light sensitive
shift registers sensor area
1.16: Output of charges with the line camera
integration times, which in turn require high light intensities of the illumination
units.
Monochrome area scan cameras consist of a matrix of CCD sensors. These have
been described in the previous sections.
Color cameras produce a color image, which consists of three parts: red, green and
blue. By additive color mixture and intensity variations in the different parts, al-
Copyright 2000 A. Erhardt-Ferron
1.2 Imaging Methods and Sensor Systems 21
most any color can be reproduced.
In the less sophisticated kinds of cameras (single chip color cameras) the
incoming light gets separated into its red, green and blue parts by means
of a stripe or mosaic filter which is located directly on the CCD sensor
(fig. 1.17). The filter stripes or the equivalent elements of the mosaic filter
are transparent for one of the three colors, respectively. At readout time, the
pixels of the red, green, and blue sectors are transferred successively. Elec-
tronic switches divide the signal into the primary colors. The mosaic filter
causes the resolution to be reduced by a factor of two in two directions; the
stripe filter reduces the resolution by a factor of three in one direction, com-
pared to a monochrome camera. Although the colors can be represented
on a TV screen, the three primary colors cannot be separated for individ-
ual image processing purposes. Single chip cameras are therefore not very
well suited for image processing tasks but are mostly used in entertainment
electronics areas.
R G R G R G R G R R R R R R R R
G B G B G B G B G G G G G G G G
R G R G R G R G B B B B B B B B
G B G B G B G B R R R R R R R R
R G R G R G R G G G G G G G G G
G B G B G B G B B B B B B B B B
R G R G R G R G R R R R R R R R
G B G B G B G B G G G G G G G G
ab
1.17: A one chip camera produces colors through
a) mosaic or
b) stripe filters
Three chip color cameras use a separate CCD sensor for each of the three
primary colors. Prisms in the optical path provide a separation of the incom-
ing light into its three components, which will be directed to the appropriate
sensor. The data of the three CCD sensors are directed towards different ar-
eas of the image RAM and can be processesed separately (fig. 1.18)
Infrared cameras are only sensitive to certain frequency bands of the infrared radia-
tion. As everybody should know, each physical body with a temperature above
absolute zero emits radiaton. The relevant physical laws are Planck s radiaton
law
8 h
u( T) = (1.2)
hc
3
kT
e 1
with:
u( T): Spectral radiation energy density
Copyright 2000 A. Erhardt-Ferron
22 1 The Image Processing System
monochrome CCD
R G B
prisms
incoming light
1.18: The incoming light is divided by prisms into its basic components red, green, and blue.
h =6:6262 10 34 Js: Planck s constant
: wavelength of light
T: Temperature in Kelvin
k =1:38066 10 23 J/K: Boltzmann s constant
c =299:8 106 m/s: speed of light in vacuum
1eV = 1:60219 10 19J: conversion factor
and Wien s displacement law, which states, where the maximum of the radiation
energy density in fig. 1.19 is located:
b
= (1.3)
max
T
with:
T: Temperature in Kelvin
b =2:8978 10 3 m K: Wien s constant
For example, a physical body at room temperature (300 K) emits infrared radiaton
with a wavelength of 10 m (IR-C) (fig. 1.14). Infrared cameras are able to
capture this radiation and, through the same principles as monochrome cameras,
transfer it into electronic signals. This is called thermography. According to
the norming commission Commission Internationale de l Eclairage (CIE) the
infrared spectrum is subdivided into three bands (fig. 1.14). While the atmosphere
is mostly opaque for infrared radiation, there are five narrow bands in the infrared
spectrum for which the atmosphere is transparent (tab. 1.4). They are used for
astrophysical tasks.
The image sensor of an infrared camera consists of semiconductor materials
which are sensitive to the infrared spectrum, i.e. for wavelengths of 0.78 m
Copyright 2000 A. Erhardt-Ferron
1.2 Imaging Methods and Sensor Systems 23
3
Energy density/mm
350000
T = 6000K
T = 5500K
300000
T = 5000K
250000
200000
150000
100000
50000
0
0 500 1000 1500 2000 2500 3000 wavelength [nm]
a
3
Energy density/mm
0.12
T = 310K
T = 277K
0.10 T = 250K
0.08
0.06
0.04
0.02
0
0 10000 20000 30000 40000 wavelength [nm]
b
1.19: The energy distribution of a black body
a) for sun-like temperatures
b) a human body (310 K), the milk in the refridgerator (277 K) and a frozen chicken (250 K)
Table 1.4: IR transparent bands in the atmosphere
Name wave length
I - channel 1:25 m
H - channel 1:65 m
K- channel 2:2 m
L - channel 3:6 m
M- channel 4:7 m
Copyright 2000 A. Erhardt-Ferron
24 1 The Image Processing System
or higher. To keep the noise level caused by the ambient temperature low, the
CCD chip of this type of camera has to be permanently cooled. Modern cooling
methods are thermoelectric cooling (Peltier cooling) and Stirling cooling. This is
the reason that infrared cameras are larger than equivalent cameras for the visible
electromagnetic spectrum.
1.2.2 Modern Camera Developments
CCD sensors still have big disadvantages in spite of the progress made in this technol-
ogy. One of them is the effect called blooming. It results from a local overexposure of
pixels to light, for example when a light source is photographed in front of a dark back-
ground. The pixels get oversaturated and the charges leak into neighboring pixels. The
result is a light spot in the image, which enlarges uncontrollably (therefore the expres-
sion blooming) and erases the actual image information. Modern detectors incorporate
an anti-blooming circuit, which prevents the leakage of charges into neighboring pixels.
However, this camera type is about 30% less sensitive.
A further disadvantage, which all the previously mentioned architectures share, is the
bottleneck of the serial readout registers, which slows down the transfer rate immensely.
Currently industry and research facilities are investing lots of money and energy into
new camera concepts - with different results. Let s have a closer look at three examples.
CMOS Technology
A big change in the digital camera market could come from the introduction of a new
type of sensor in CMOS cameras (Complementary Metal-Oxide- Semiconductor).
For CMOS sensors the production process is the same as for all micro processors,
storage chips and ASICs (engl. Application Specific Integrated Circuit). With CCD -
cameras incoming photons create negative charge carriers, which have to be integrated
over a certain time which is determined by the shutter speed. By contrast, CMOS -
Sensors are, in principle, photo sensitive diodes, which are in series with a resistor.
While in CCD - cameras electrons have to be transported via shift registers, the principle
of the CMOS Camera allows a continuous transformation of the incoming photons into
a resulting voltage. A CMOS - sensor is nothing but an array of exposure meters (fig.
1.20.
In the past, CMOS sensors were not very popular because of their high noise level,
which was caused by the layout of the chip. However, in 1993 physicists at the NASA
Jet Propulsion Laboratory in California succeeded in developing a unit called APS (Ac-
tive Pixel Sensor), which incorporates, for each pixel, an active transistor amplification
circuit, providing, among other things, efficient noise reduction. In spite of that, the
signal - to - noise ratio of a CMOS - camera is still significantly higher than that of a
CCD+- camera. Here are some basic features of CMOS - Chips:
Direct access: CMOS - cameras permit direct access on any pixel, just like computer
storage units. Like the image buffer, it can be addressed via line- and column
coding (see. section 1.3.2).
Copyright 2000 A. Erhardt-Ferron
1.2 Imaging Methods and Sensor Systems 25
photo detector
active amplifier
1.20: CMOS - array with APS. Every pixel can be addressed individually through line and column
selection
All image processing functins on one chip: VLSI - technology (Very Large Scale In-
tegration) permits the integration of all necessary camara functions directly on
the CMOS Chip. In addition, it is possible to add more intelligent signal process-
ing circuits to provide, for example, image compression, image optimizing, color
coding, segmentation etc. [22]. In principle, with CMOS cameras, all image
processing algorithms, which will be addressed in the following chapters, can be
implemented directly on the camera chip.
Wide dynamic range: Because of special features of the CMOS structure[11] the trans-
formation process of photons into voltage is not linear but logarithmic, i.e. actu-
ally similar to the light response of the human eye. While a CCD camera has a
dynamic range of about 2 - 3 decades, the range of the CMOS camera includes 6
decades and more. It permits the taking of an image of a 100 Watt light bulb and
enables the recognition of the filament (fig. 1.21).
1.21: Image of a light bulb taken with a High Dynamic Range CMOS camera
Low power consumption: A CMOS-camera uses less electricity (by a factor of 100)
than a CCD - camera. A CCD - camera will use about 2- 5Watt of power while
a CMOS - camera needs only 20 - 50 mW A NiCd camcorder battery is used up
in a few hours when connected to a CCD - camera, while it might last a week or
Copyright 2000 A. Erhardt-Ferron
26 1 The Image Processing System
more when connected to a CMOS - camera. Less power consumption makes an
image processing system more flexible. In a few years, all image processing algo-
rithms will take place on chips in intelligent cameras, which might be connected
to notebook computers.
Low costs: The production process of CMOS - wafers is much simpler than CCD pro-
duction. Therefore the price of a CMOS - camera today is not much higher than
the price of a CCD - camera for industrial applications in combination with a
frame grabber hardware[22].
No blooming: Since CMOS cameras read out every pixel individually, blooming can-
not occur.
High data transfer rates: The parallel transfer allows very fast image capturing and
processing because routing through the horizontal and vertical shift registers is
prevented. At this time the upper limit of the image transfer rate is about 1000 im-
ages per second, if an image size of 1024 1024 pixels is assumed[23].
Research is moving in the direction of development of so-called  intelligent cameras
, which will be capable of taking over computer functions. Research labs have been
working on cameras with the ability to adapt to light changes, cameras with stereo-
scopic vision, cameras with integrated capabilities of smoothing and edge detection
etc. While most cameras which are currently available are still based on the interline
transfer technology[20][22], conform to the CCIR norm and transfer images with the
interlace mode, a new camera generation will probably exceed the ability of the human
eye[15] in the near future.
Multichannel Sensors for Color Recognition
Current color cameras incorporate color filters or three separate chips for the basic col-
ors red, green and blue. Research groups, among them a group at the Research Center
in Juelich, Germany, are developing sensors, which store the coor information of a light
ray in photo sensitive layers of a pixel, which are mounted on top of each other[18] For
that task porous silicon is used[10]. It is manufactured from commercially available sil-
icon wafers, which are a widely used product in the semiconductor industry. Pores with
diameters of a few nanometers are electrochemically etched into the material, the size
of which is much lower than the wave lengths of visible light. This means that a porous
layer will have a refractive index which depends on the size of the pores. By controlling
the density of the electical current during the production process, layers with different
refraction indices can be created. When white light reaches the silicon material with
multiple refraction layers, certain colors will be absorbed in certain layers, while the
rest of the spectrum is able to transfer to lower layers. After the integration time, the
color information of the incoming light will be stored vertically in multiple layers of
the pixel. Fig. 1.22a) shows a cross section through a photo diode with integrated lay-
ers of porous silicon. Fig. 1.22b) shows the spectral responsiveness of a sensor with
three colors as a function of the wavelength. So far the research groups have succeeded
Copyright 2000 A. Erhardt-Ferron
1.2 Imaging Methods and Sensor Systems 27
Cr/Au
PS
depletion
p+-Si
zone
200
150
Iph
n--Si
100
Iph
50
0
300 400 500 600 700 800
contact
wavelength [nm]
a b
1.22: A porous silicon photo sensor[18]
a) Cross section
b) Spectral response of a porous silicon photo sensor with three layers
in developing a wafer with six different layers[18]. Theoretically, it should be pos-
sible to vary the refraction index continuously with the thickness of the silicon. The
wavelengths which can be covered by this method are said to range from 300 nm to
3000 nm.
Intelligent Thin Film on ASIC (TFA) Imaging Sensors
With the CCD - Sensor as well as with the CMOS - Sensor not 100% of the chip
surface is light-sensitive, since both sensor types have to integrate storing and readout
areas in between the photo-elements. TFA (Thin Film on ASIC) technology overcomes
this drawback with the integration of an amorphous silicon detector (a-Si:H) on top of
a crystalline ASIC (Application Specific Integrated Circuit)[6] (fig. 1.2.2).
The layer of amorphous silicon is less than 1 m thick and acts like a multispectral
photo-diode. The maximum of the spectral response can be shifted across the visible
spectrum from red on one end to blue on the other end by changing the voltage between
p- and n-layer of the diode. It is the only electronic element worldwide which shows
three linear-independent sensitivity peaks in the visible part of the spectrum, which
are linear to the light intensity, and from which the RGB color signal can be derived.
On the ASIC layer, each pixel is connected to circuitry which sequentially applies the
three voltages to the diode. This way all three basic colors necessary for additive color
composition are realized in one pixel and are read out one after another.
Because of the programmable ASIC, multiple readout procedures can be realized. It
is possible, for example, to read only the red, blue or green part of an image, to read
the RGB information of single lines or columns or predefined pixels (random access).
In addition, pixels can have circuitry with functions required by the customer. In the
Copyright 2000 A. Erhardt-Ferron
spectral sensitivity [mV]
28 1 The Image Processing System
Front electrode
a-Si:H thin film system
}
Rear electrode
a-Si:H-Detector
Insulation layer
ASIC
ASIC
a b
1.23: Photosensor in TFA-technology[6]
a) The principle of a TFA Sensor
b) The layers
most simple case it will be one or more MOS - transistors (Metal Oxide Semiconduc-
tor) for parallel readout or random access on pixel addresses. But also more complex
functions may be integrated, so that in principle image processing algorithms like data
compression or application specific intelligent procedures required by the customer can
be programmed right into the ASIC.
Since all electronic circuitry is realized in the ASIC, the total sensor surface is light
sensitive, which means a so-called filling factor of 100%. Like some of the other new
camera developments TFA cameras are still under development and cannot yet be pur-
chased. There are only a few prototypes and the future will show if they can beat the
competition of the camera market.
1.2.3 Camera Lenses
A camera lens consists of a lens system and one or more apertures. The lens system
and the aperture control the amount of light which will reach the sensor as well as the
depth of field. A small iris results in a large depth of field, but also causes undesired
diffraction effects. A large iris results in blurred pictures when the object extends into
the third dimension.
Optical Basics
An object is projected onto a sensor through an optical system consisting of mirrors and
lenses. In addition there is an iris which controls the incoming amount of light and the
depth of field of an image, i.e. the range in front of and behind the object, which will
appear in focus.
Copyright 2000 A. Erhardt-Ferron
Optical Detector
1.2 Imaging Methods and Sensor Systems 29
The amount of light which passes through the lens system and ultimately reaches the
sensor is proportional to the size of the aperture and the exposure time.
The laws in this chapter are so-called thin lens approximations; for thicker lenses,
the formulas are much more complicated because aberrations and other lens inconsis-
tencies influence the image depending on the thickness of the lens. But modern lens
systems also incorporate several corrective lenses to eliminate geometric and chromatic
distortions. Therefore, for practical purposes the assumption of thin lenses can be made
in conditions where the distance of the object is at least ten times the focal length.
light source focal point
f f
optical axis
"
a
light source
focal point
f f
optical axis
b
1.24: Imaging through a lens:
a) a light source in an infinite distance is focused on the focal point
b) if the light source moves closer, the light rays collect behind the focal point
Fig. 1.24 shows the geometric principles of a lens. Rays which originate from a light
source at an infinite distance are parallel. A lens which is positioned perpendicularly to
these light rays will focus them at the focal point. This means that the focal point is the
image of a light source at an infinite distance. The distance between the center of the
lens and the focal point is called the focal length f. This is also the distance between
the lens and the CCD chip if an object in an infinite distance is to be captured. Then
the chip is said to be located in the focal plane. If the light source gets moved towards
the lens, the rays will be focused behind the focal point. Consequently, the distance
between the lens and the CCD chip has to be enlarged. The relevant laws for thin lenses
were compiled by Descartes and combined in his famous lens equations (fig. 1.25):
Copyright 2000 A. Erhardt-Ferron
30 1 The Image Processing System
G
f f B
g b
1.25: The optics of a thin lens
1 1 1
+ = (1.4)
g b f
with
b: distance of the image
g: distance of the object
f: focal length
and
B b
= = m (1.5)
G g
with
B: height of the image
G: height of the object m: aspect ratio
This indicates that focusing means simply changing the distance between the lens
and the CCD chip. Obviously, however, this will eventually encounter geometric limits.
Normally, a lens system allows the focusing from an infinitely distant point up to the
so-called minimum object distance (MOD). the size of which follows from equation 1.4
if b = bmax and g = MOD:
1 1 1
+ =
bmax MOD f
f bmax
! MOD = (1.6)
bmax f
with
bmax: maximum distance of the image
MOD: minimum distance of the object
f: focal length
The minimum and maximum object distance can be reduced by a spacer ring, which
reduces the distance between the lens system and the camera chip.
Before a lens is purchased, an estimate of the focal length f has to be calculated,
taking into account the geometric facts at the site. It can be derived from 1.4 and 1.5:
1.5:
mg
f = (1.7)
1 + m
b
= (1.8)
1 + m
Copyright 2000 A. Erhardt-Ferron
1.2 Imaging Methods and Sensor Systems 31
with
f: focal length
g: distance of the object
m: aspect ratio
Furthermore, from equations 1.4 and 1.5 one gets the useful relations:
b = f (1 + m) (1.9)
1
g = f 1 + (1.10)
m
with
f: focal length
g: distance of the object
b: distance of the image
m: aspect ratio
The shorter the focal length, the greater the refractive power D of the lens. The
refractive power D is the inverse of the focal length f:
1
D = (1.11)
f
People who wear glasses are familiar with D, which is called the diopter and is measured
in 1/m.
In addition to the focal length, the photographic angle # is another characteristic of a
lens system (fig. 1.26).
CCD chip
light source focal point
Ń
f f
optical axis
"
1.26: The photographic angle #
It is defined as
Bmax #
= tan
2f 2
Bmax
! # = 2 arctan (1.12)
2f
Copyright 2000 A. Erhardt-Ferron
32 1 The Image Processing System
where Bmax is the length of the diagonal of the chip in the case of area scan cameras and
the width of the line in the case of line cameras. With one lens numerous photographic
angles are possible, if different chip sizes are considered (tab. 1.5 p. 36).
As mentioned previously, the thicker a lens, the more the optics will differ from
the above-made assumptions of a thin lens, and the greater the distortions will be. In
addition, the thicker a lens, the smaller the focal length f and the larger the photographic
angle. It can be calculated that with C-mount lenses, the distortions will start at a focal
length of f=8 mm. This value can still be improved if highly refractive glass is used.
Therefore, in metrology, C-mount lenses with focal lengths of less than 8 mm should
only be used in exceptional cases since the correctional calculations which the image
processing system will have to perform could be very time consuming and therefore
expensive.
Another important parameter is the depth of field which has been previously men-
tioned. Fig. 1.27
x y
2r 2R
f f
gl gr
CCD chip
pixel
g (image plane)
b
1.27: Field of depth calculation
shows the origin of the effect. If a CCD chip shows a focused image of a point which
is located at the object distance g (straight line), the image of a point which is located
at gr (dashed line) or at gl (points) will be an unfocused circle. If the diameters of these
two circles are smaller than the length of the edge of a pixel, any object which is located
between g + gl andg- gr will be in focus. The distance
g + gl (g gr ) = gl + gr (1.13)
is called depth of field. It depends on the size of the iris of the lens along with other
physical parameters. A small iris enlarges the depth of field; a larger iris makes it
smaller. As you might know, photographers usually use a large iris for portrait shots in
order to blur the background and to emphasize the face completely. The size of the iris
is measured in f-stops k, which are well-known among photographers. In mirror reflex
p
cameras it can be set in steps of 2: k = 0.71, 1, 1.4, 2.0, 2.8 etc. A video camera
makes a continuous setting possible and the motor can be controlled by software. The
f-stop is defined by the following equation:
f
k = (1.14)
2R
Copyright 2000 A. Erhardt-Ferron
1.2 Imaging Methods and Sensor Systems 33
where f is the focal distance and R the effective radius of the iris. In the case of the
combination of the iris and a thin lens, R is equal to the radius of the iris. With thicker
lenses the f-stop concept does not hold, and the opening of the lens- iris system is
measured by the quantity called numerical aperture N.A.:
#
N:A: = n sin (1.15)
2
where n denotes the refractive index of the lens environment and # is the photographic
angle, defined in equation 1.12.
If we stick with thin lenses in this context, a small radius R of the iris amounts to
a large f-stop k and vice versa. The f-stop controls the amount of light falling on the
p
sensor (as can easily be calculated enlarging the f-stop by a factor of 2 will reduce
the area of the iris and therefore the amount of light by a factor of 2) as well as the depth
of field. Camera lenses take as a measure of the light intensity the inverse of the f-stop
number, the so- called relative pupil diameter
1 2R
= = (1.16)
k f
with
k: f-stop number
f: focal length
R effektive radius of the pupil.
If you have, for example, a 50 mm lens which has the ratio 1:2.8 imprinted on its body
it means that the diameter of the pupil amounts to 50 mm/2.8 = 17,9 mm.
Descartes s lens equation applied to fig. 1.27 results in:
1 1 1
+ = (see equ.1.4)
g b f
1 1 1
+ = (1.17)
g + gl b x f
1 1 1
+ = (1.18)
g gr b + y f
In addition, the following equations hold:
R r
= (1.19)
b x x
R r
= (1.20)
b + y y
from which follows
rb
x =
R + r
rb
y =
R r
If x and y are inserted in equations 1.17 and 1.18 and the equations are solved for g
l
and gr respectively, considering equations 1.4 and 1.14, the result is the range within
Copyright 2000 A. Erhardt-Ferron
34 1 The Image Processing System
which the object will be in sharp focus.
2rk g(g f )
gl = and
2
f 2rk(g f)
2rk g(g f )
gr =
2
f +2rk(g f)
2
4f rk g(g f)
! gr + gl = (1.21)
4 2
f 4r2k (g f)2
Please note that gr and gl do not have the same magnitude (fig. 1.28). From the above
equations, it follows that the depth of field for a given CCD chip with a given pixel size
depends on the object distance g (fig. 1.28 a), the f-stop k (fig. 1.28 b), and the focal
length f (fig. 1.28 c). If the focal length f and the f-stop k are constant and the camera
is moved away from the object as in fig. 1.28 a, there will come a point, when g will
l
reach infinity. This is the case when the denominator of equation 1.21 approaches zero,
i.e. when
2
f = 2rk(g f )
2
f
! g = + f (1.22)
2rk
g
This distance is called the hyperfocal distance. Then gr is exactly .
2
Example: The pixels of a CCD camera have a size of 16 m 16 m ; the camera
is equipped with a 50 mm lens, and the distance between an object and the camera
is g = 1 m. The camera is set to an f-stop of k = 8. From equation 1.21 this results in a
depth of field of gl + gr = 97.5 mm. The hyperfocal distance amounts to g = 19.58 m.
Another way to express equation 1.21 can be obtained by using equations 1.4 and 1.5:
2rk f(m +1)
gl =
m(fm 2rk
und
2rk f(m +1)
gl =
m(fm +2rk
bzw.
2
4f rk(m +1
gr + gl = (1.23)
2 2
f m2 4r2 k
with
m: aspect ratio
The aspect ratio causing the hyperfocal case would be
2rk
m = (1.24)
f
Lens Types
Camera lenses are divided in differnt categories like wide angle lenses normal lenses
and telephoto lenses. These subdivisions have historical origins, though, when cameras
Copyright 2000 A. Erhardt-Ferron
1.2 Imaging Methods and Sensor Systems 35
focal depth [m]
10000
1000
gr
gl
100
g +g
r l
10
1
0.1
0.01
2 4 6 8 10 12 14 16 18 g [m]
a
focal depth [cm]
25
20
gr
gl
g +g
15 r l
10
5
0
2 4 6 8 10 12 14 16 k
b
focal depth [cm]
70
60
gr
50
gl
g +g
r l
40
30
20
10
0
20 40 60 80 100 120 140 f [mm]
c
1.28: The field of depth as a function of g, k and f respectively (equ. 1.21. r = 8 m)
a) as a function of the object distance g. f = 50 mm, k=8
b) as a function of the f-stop k. g=1m, f =50 mm
c) as a function of the focal length f. k = 8, g=1m
Copyright 2000 A. Erhardt-Ferron
36 1 The Image Processing System
used 35 mm - film. The categories would have to be modified as soon as a new CCD
chip hits the market. But this, of course, is unrealistic. Instead, the terms wide angle
lens, normal lens and tele lens still to this day relate to the 35 mm - film with a frame size
of 24 36 mm. Given this frame size, the diagonal is 43.3 mm. With a photographic
angle of # =45 , which is roughly equivalent to the human field of vision, equation 1.12
and tab. 1.5 p. 36 yield a focal length of approximately f =50 mm. Therefore this lens
is called normal lens. All camera lenses with a larger photographic angle are called
wide angle lenses, and the ones with a smaller photographic angle are called telephoto
lenses. If the CCD - camera chip is small enough, even a 50 mm - lens may have the
same photographic angle as a telephoto lens. This is an important factor which has to
be considered when selecting a camera lens.
Table 1.5: Photographic angles and lens names
Format 24 36 mm 1" 2/3" 1/2" 1/3" 1/4"
Diagonal Bmax [mm] 43.3 15.9 11.0 8.0 6.0 4.0
f = 20 mm 95 43 31 23 17 11
f = 24 mm 84 37 26 19 14 10
f = 35 mm 63 26 18 13 10 7
f = 50 mm (Normal) 47 18 13 9 7 5
f = 105 mm 23 9 6 4 3 2
f = 135 mm 18 7 5 3 3 2
f = 180 mm 14 5 4 3 2 1
f = 300 mm 8 3 2 2 1 1
Commercially available lenses incorporate lens systems for aberration corrections
(fig. 1.29
The Tessar - lens, the Gaussian twin lens, the Cooke triplett and the Petzval lens
are usually constructed as 50 mm lenses
Wide angle lenses (6 mm bis 40 mm) for example the Aviogon- or Orthogometer -
lens, have smaller focal lengths and a large photographic angle.
Telephoto lenses like the Magnar lens, have large focal lengths and a small pho-
tographic angle.
Zoom lenses with variable focal lengths are of course the most complicated chal-
lenges in lens design.
In addition to the basic formulas derived in the previous section there are, in reality,
other parameters to be considered when a lens system has to be integrated into an image
processing application. Different imaging tasks require different lens types in addition
to which filters and front lenses might have to be integrated.
C and CS mount lenses: C-mount and CS-mount lenses are attached to the camera
body by means of a threaded connection. These two types of lenses differ only
Copyright 2000 A. Erhardt-Ferron
1.2 Imaging Methods and Sensor Systems 37
a b
c d
e f
g
1.29: Common lens types[19]
a) Wild Aviogon Lens b) Gaussian Twin Lens (Biotar)
c) Zeiss Orthogometer Lens d) Cooke (Taylor) Triplet
e) Tessar Lens f) Petzval Lens
g) Magnar Tele lens
Copyright 2000 A. Erhardt-Ferron
38 1 The Image Processing System
in the distance between the thread and the focal plane. The C-mount lens has a
distance of 12.5 mm; the CS-mount lens of 17.5 mm. If a spacer ring is added to
a C-mount lens, it will become a CS-mount lens. Similarly to chip formats, the
origin of C mount lenses goes back to the age of tube cameras. Typical sizes are
1/3 inch, 1/2 inch, 2/3 inch, and 1 inch. In general, the format of the lens should
be larger than or equal to the format of the CCD chip, so that the edge of the iris
will not show up on the picture. Also, aberrations appear mostly at the edge of a
lens. Consequently, a CCD chip format of, for example, 1/4 inch requires a lens
format of 1/3 inch.
In addition to lenses with fixed focal lengths, manual and automatic zoom lenses
are available, as are lenses on which the f-stop and focal length can be controlled
by the video signal. In addition, all the lenses still available on the camera market
can be used with adapters, which connect bayonet joints to C-mount joints.
Macro lenses: Macro lenses have to be used if the object is located at such a short
distance from the camera that spacer rings and macro auxiliary lenses can no
longer meet the requirements. This is the case when the aspect ratio m is be-
tween 0.1 and 10. They are specifically laid out for metrology, i.e. anything
which might contribute to the instablility of the optical system, e.g. a changeable
iris or changeable focal length is omitted. Therefore the light conditions and the
mechanical setup have to meet the requirements of the macro lens. While normal
lenses have the focal length as a typical parameter, macro lenses are differentiated
by the aspect ratio.
Telecentric lenses: If a three-dimensional object is photographed, the image usually
also contains the depth of a scene (fig. 1.30 a)). With some tasks, however, this
is undesirable. In such cases, telecentric lenses are used. They have a second
small iris, which is located directly in the focal point on the opposite side of the
camera body. Then only light rays which are mostly parallel to the optical axis
will end up on the CCD chip, i.e. light rays coming from surfaces which are
exactly perpendicular to the optical axis (fig. 1.30 b)). However, because of the
small second aperture, only the light rays through the center of the lens will be
captured on the sensor. Therefore only objects which are either very far away or
considerably smaller than the diameter of the lens can be captured.
If the object is moved along the optical axis, the image size should not change
with this type of lens but in reality, this is only possible within certain limits.
Therefore, a telecentric lens has a quality parameter called the telecentric range.
If the object is moved along the optical axis within the telecentric range, the
image size will change less than 1 mm.
Lens Add-Ons
In addition to lenses there are various add-ons which make taking images easier.
Auxiliary lenses: Macro auxiliary lenses and filters can be attached to the lens at the
opposite side of the camera body. They have the same result as the attachment of
Copyright 2000 A. Erhardt-Ferron
1.2 Imaging Methods and Sensor Systems 39
a b
1.30: Telecentric lenses
a) mapping through a normal lens
b) mapping with a telecentric lens
spacer rings, i.e. to lower the object distance g. They are usually attached with
zoom lenses, where a spacer ring would result in mechanical instability. Fig. 1.31
explains the principle. Without a macro auxiliary lens, a point which is located
at an infinite distance will be captured in the focal plane, as mentioned above
(fig. 1.31 a)). If we attach a macro auxiliary lens, an object in the focal plane of
that lens will have an image in the focal plane of the main lens as is shown in
fig. 1.31 b). If the object is moved, an object distance g has to be set at the scale
of the main lens. This scale, however, has been defined for the main lens alone,
and not for a system in combination with the auxiliary lens. The new setting of
the object distance g can be derived from the following formula:
1 1 1
= + (1.25)
fNew f fN
1 1 1
= + for d < < g (1.26)
f b g d
with:
f: focal distance of main lens
fN : focal distance of auxiliary lens
fNew: common focal distance of main and auxiliary lens
b: distance of the image
g: distance of the object according to the scale of the main lens
d: distance between auxiliary and main lens, d < < g
g d g: distance between auxiliary lens and object
From basic optics we know that the refraction power of a system of more than
one lens is the sum of the refraction power of each lens. Using Descartes s lens
equation, we get:
1 1 1
+ =
g0 b fNew
1 1
= +
f fN
Copyright 2000 A. Erhardt-Ferron
40 1 The Image Processing System
light source focal point
f f
optical axis
"
a
light source
focal point focal point
closeup lens
f
f
N
optical axis
d
b
1.31: A closeup lens and a normal lens
a) without closeup lens: parallel lightrays are focused on the focal point f
b)lightrays from the focal point of the closeup lens f are focused on the focal point f
N
1 1 1
= + +
b g d fN
1 1 1
=
g d g0 fN
g0
g d = fN
fN g0
g0
=
1 g0DN
g0
! g for d << g (1.27)
1 g0DN
with:
g0: real distance of the object
DN = 1=fN: refractive power of the auxiliary lens g: distance of the object
according to the scale of the main lens.
For example, let s mount a macro auxiliary lens with a refractive power of 3 diopters.
If the object has a distance from the camera of 30 cm, the scale on the main lens
has to be set to 3 m.
Spacer rings: As mentioned above, the MOD can be reduced by including a spacer
Copyright 2000 A. Erhardt-Ferron
1.3 The Frame Grabber 41
1.32: Spacer rings of various sizes
ring (fig. 1.32) between the lens system and the camera body to enlarge the dis-
tance between the lens and the camera chip. However, as a result, focusing on
objects at a great distance will no longer be possible the larger the spacer ring,
the smaller the minimum and maximum object distances.
Polarization filter: Polarisation filters are well-known to hobby photographers as a
useful tool to prevent reflexions from shiny surfaces in a picture. Light rays
which are reflected by a mirror or even a shiny surface have a certain direction of
polarization, i.e. the wiggling of the light waves is limited to a a certain plane.
A polarization filter which, like an auxiliary lens, is attached to the lens on the
opposite side of the camera body, can be turned in a way that it will absorb the
light of a certain polarization plane and therefore prevents it from getting to the
CCD-chip.
Of course there are lots of add-ons on the market to treat images artistically, for exam-
ple, color filters and various effect filters. However, they are beyond the frame of our
interests and will not be dealt with here.
1.3 The Frame Grabber
The electical voltage signal produced by the sensor system will now be transferred to
the frame grabber. The frame grabber is not identical to the graphic card in normal
computers. It has to meet many more requirements.
A frame grabber should be able to
process the information from various image sources
store image information quickly and efficiently
offer a graphics user interface (GUI)
be flexible concerning various applications
Copyright 2000 A. Erhardt-Ferron
42 1 The Image Processing System
Depending on the type and the price a user is willing to pay, a frame grabber might
include fast DSPs with RISC architectures and multiple processor systems for parallel
processing, large RAM storage capacities, sophisticated software libraries, interactive
user interfaces and comfortable programming tools. Nowadays the frame grabber board
is still the main component in an image processing system, although with the increasing
availability of CMOS sensors, the image processing routines will be increasingly moved
directly into the camera (see section 1.2.1). On the other hand, image processing algo-
rithms can be programmed directly in the host computer. Time-critical applications,
however, require a hardware frame grabber unit, either in the camera or as a hardware
unit added to the host PC or workstation.
To meet the numerous requirements, most frame grabbers have a modular structure,
so that the components can be configured according to the needs of the user.
Modern frame grabbers usually consist of the components (fig. 1.33):
video input unit (VIU)
frame buffer (FB)
digital signal processors (DSP)
video output unit (VOU)
DSP
VIU VOU
FB
1.33: Hardware components of an image processing
system: video input unit(VIU), frame buffer (FB),
signal processors (DSP) and video output (VOU).
The frame buffer, as well as various digital signal processors, will be called the image
processing unit. Since the market offers a very large spectrum of frame grabber types,
it is difficult to describe a typical structure. The following sections can therefore only
give an overwiew.
1.3.1 The Video Input Unit
The video input unit is the interface between the sensor system (i.e. a CCD camera)
and the image storage unit.
Like cameras, frame grabbers offer several features which exceed the video norm.
Some types can be connected to almost any signal or image source (fig. 1.34). Basically,
the different kinds of data sources can be named:
analog normed data (from video cameras, video recorders, etc.),
Copyright 2000 A. Erhardt-Ferron
1.3 The Frame Grabber 43
analog
normed
video camera
video data
video recorder
raster
elektron analog unnormed
VIU
microscope
image data
scanner
CD-ROM
digital image data
tape
1.34: Various image and data sources
analog unnormed data (from computer tomographs, electron microscopes, line
cameras, etc.)
digital data (from CDs, CMOS image sensors, etc.)
Because of the required flexibility, a frame grabber board has to be configurable by the
user. With any of these models, the video input unit has to be able to
multiplex the input sources
synchronize the incoming signal with the RAM
digitize analog data
transfer digital data
pre-process data
parallel/serial
interface
input
lookup
multiplexer AD converter table
sync.
external
separation
trigger
1.35: Functional units of the video input.
Therefore the following functional groups are required:
Copyright 2000 A. Erhardt-Ferron
image acquisition
image storage
44 1 The Image Processing System
a multiplexer: Often the image information consists of multichannel video signals
which are connected to the different inputs of the frame grabber board, for ex-
ample the red, green and blue channels of a three chip color camera, a system
of multiple black and white cameras or satellite data consisting of five and more
channels.
Boards at the high end of the performance range are able to read and process
all channels in parallel mode. Boards in the middle and low end of the range
have a multiplexer in the entrance stage, by which a video source can be selected.
The signal selection can be controlled by software. In order to be able to switch
between the various video sources, they have to be externally synchronized by
the PC clock or a sync generator on the frame grabber board. This unit is part of
sync separation.
sync separation (also: Sync Stripper): According to table 1.1 video signals contain the
image information plus sync signals for line start and field start. The horizontal
sync signal introduces a new line; the vertical sync signal the beginning of a new
field. The sync separation unit removes these two signals from the image data.
analog digital converter: The AD converter digitizes the analog input signal. The
resolution of standard AD converters is 8 bit. Some boards have variable AD
frequencies which can be changed by software. This makes the input of signals
which do not conform to any of the video standards possible. These systems are
called variable scan systems. With these systems the video input unit contains an
additional trigger input (variable scan input) through which it receives the trigger
signal from the video source. (fig. 1.35).
parallel and serial interface: Some images are already digitized when they reach the
frame grabber board, either because the image sensor is able to produce digital
data or because the data is pre- processed in the image capturing system. In
such cases, the video input unit provides digital interfaces which omit the AD
converter.
input lookup table: Before the data is transferred to the frame buffer, some boards al-
low the possibility of adding modifications via an electronic transformation table.
Input and output lookup tables (section 1.3.4) are a number of addional memory
areas. They can be accessed via the GUI and filled with numbers. The value from
the AD converter is interpreted as a relative memory address. The value behind
this address will then be transferred to the frame buffer (fig. 1.36). The user is
able to select a lookup table in real time. This makes it possible to set a threshold
or to, for example, eliminate undesired parts of an image before the data reaches
the frame buffer.
1.3.2 The Frame Buffer
Data can be stored either in the frame buffer of the image processing board or in the
PC s local storage. Modern PCI-bus computers provide sufficiently high data transfer
Copyright 2000 A. Erhardt-Ferron
1.3 The Frame Grabber 45
255
254
253
.
.
.
) ) ) )
A/D 94 70 image storage
94 70
.
.
.
2
1
0
LUT
1.36: Input lookup table
bits
frame 2
(0,0)
frame 1
frame 1 frame 3
(0,0)
frame 2
frame 3
(0,0)
image storage
image storage
a b
1.37: Configuration possibilities of the frame buffer: images with different
a) sizes
b) and depths
Copyright 2000 A. Erhardt-Ferron
46 1 The Image Processing System
rates to reach an acceptable processing speed. If the PC itself is equipped with sufficient
storage space, it s even possible to do without the frame buffer. Indeed, in time-critical
tasks, all processing steps, including the storage of the image, take place in the image
processing board.
Regardless of where the frame buffer is located, it has a different administrative struc-
ture and access possibilities than normal RAM storage. While normal PC RAM is
continuously addressed, with the frame buffer, the user has the impression of working
on a a matrix with x- and y- coordinates. The conversion from continuous to two-
dimensional addressing and to the configurations described below is peformed by the
image processing system s program library.
Therefore, we define a frame buffer as a RAM area (which can be physically mani-
fested anywhere in the system) in connection with a library, which, among other things,
administers the addresses.










bit layer 0

bit layer 1





.
.

.

bit layer 7

1.38: 8 bit-planes of a monochrome image
It allows the possibility of configuring the frame buffer freely to store images
with different sizes and depths. A frame buffer of 1 MByte can, for example, be
used to store an image with dimensions of 1024 1024 pixels and 8 bit depth,
but also for a real-color image of 3 512 512 pixels (and 512 512 bytes to
store intermediate results). A pixel in a real-color image is addressed the same
way as pixels from monochromatic images, i.e. with a storage address (x ,y0)
0
although in reality it consists of three bytes (one each for R, G, and B) which
might be stored physically in totally different parts of the buffer. A 1 MByte
frame buffer might also be used for an image file of 512 1024 pixels and
additional overlay levels (up to 8 bit) to display text or marks from the mouse,
or to store an image sequence of 256 images, 64 64 pixels each. If the image
processing system uses a CMOS camera (section 1.2.2) the image buffer must
have a pixel depth of 20 pixels because of the large dynamic range of this camera
type. The image buffer of 1 MByte will in this case be able to hold two images
of 512 512 pixels each. However, in any of these configurations the user will
want to address an image like a two- or, in the case of an image sequence, a
three-dimensional matrix without having to think about pixel depth or addressing
algorithms (Abb. 1.37).
Copyright 2000 A. Erhardt-Ferron
1.3 The Frame Grabber 47
It provides several access modes. For example, an image line or column can
be addressed with a single command; likewise, bit levels can be individually
addressed.
The concept of a dual-ported memory makes it possible to address the frame
buffer in parallel from two sides. As a result, images can be read into the frame
grabber and, at the same time, be displayed on a monitor.
1.3.3 The Signal Processor
As mentioned previously, high end performance frame grabbers have the possibility
of capturing data from various input channels in parallel. Processing this amount of
data can only be performed by customized signal processors. Therefore, some frame
grabbers contain one or several DSPs, for example the model TMS320C80 (Texas In-
struments), which is a 32 bit DSP, to perform complex image processing algorithms
like filters, convolutions, data transforms or data compression. For special tasks like
neighborhood operations, specific ASIC modules can be customized. As an example,
a 3 3 convolution of an image with 512 lines and 512 columns takes about 1.8 ms,
a convolution over 5 5 neighborhoods of the same image will take 4.8 ms. These
tasks exceed the functions of a basic frame grabber; therefore, these boards are called
image processing systems. The DSPs may be located on the frame grabber board or on
separate boards, which communicate via the PCI bus or internal bus systems.
Some applications even distribute the signal processors among various computer sys-
tems and are able to communicate via LANs.
1.3.4 The Video Output Unit
The video output unit of an image processing system makes it possible to display an
image from the frame buffer on the screen. It has to transform the image data into
R
output
G
lookup-
DA converter
table
B
display processor
1.39: Functional units of the video output.
an adequate video signal. This means that the content of the frame buffer has to be
Copyright 2000 A. Erhardt-Ferron
image storage
monitor
48 1 The Image Processing System
transferred into an analog signal, which conforms to a video norm (SVGA, CCIR etc.).
The video output unit consists of two functional groups:
the output lookup table
the DA converter
As mentioned previously, lookup tables are memory areas which make it possible to
modify pixel values deliberately. Output lookup tables generally have the same prin-
255
254
253
. . .
. . . 90 )
. . .
) ) )
image storage 70 55 ) D/A
70 90 55 10
. . . )
10
. . .
. . .
2
1
0
R G B
1.40: Output lookup table
ciple as input lookup tables, although an output lookup table usually has three com-
ponents: a red, a green and a blue component. Therefore, any value from the frame
buffer can be converted into three values which provide the three basic colors neces-
sary for a color display of an image, although the image was originally captured by a
monochrome camera. This way the values of an 8 bit pixel can be transformed into 256
colors; a pixel with more bits into respectively more colors.
The digitized values from the frame buffer have to be converted into an analog signal.
This is provided by the DA converter. In addition, depending on the video norm used, a
certain number of pixels has to be transformed in a certain time frame. The time basis
information is given by the video input unit if frames are captured and displayed under
the same video norm.
New sync signals have to be generated if the input and the output video norms are
not identical. Decoupling of the input and output frequencies is done via a graphic
processor, which produces new horizonal and vertical sync signals as well as the scan
frequency for the DA converter.
As mentioned above, images in the frame buffer can differ in width and height but
also in the number of bits per pixel. This means that the image information in the frame
buffer has to be transformed according to the display parameters of the video norm
used. This is also done by the graphic processor.
Copyright 2000 A. Erhardt-Ferron
1.4 Summary 49
1.4 Summary
There are numerous types of image processing systems on the market. The one de-
scribed in this section is essentially the common denominator. An image processing
system consists of:
an illumination unit: For most applications daylight is unsuitable. Adequate illumi-
nation saves computer time. Bad illumination often leaves irreparable artefacts
in images.
a sensor unit, for example a CCD camera: The camera market is flooded with nu-
merous models. A camera purchase should be carefully considered, with an op-
timal image processing system in mind.
one or more lenses, which suit the problem: The lens format has to be greater than
or equal to the chip format. The use of spacer rings and closeup lenses can lead
to aberrations. Telecentric lenses prevent perspective distortions.
a frame grabber board: Time-critical problems require a frame grabber board with
intelligent hardware, on which fast signal processors take over most of the calcu-
lations.
suitable peripheral units for the output of results: (monitor, printer, I/O - board)
The development of image processing systems is going in a direction which will make it
possible in a few years to integrate all the hard- and software into an intelligent camera
with minimum dimensions.
Copyright 2000 A. Erhardt-Ferron
50 1 The Image Processing System
Copyright 2000 A. Erhardt-Ferron
Bibliography 51
Bibliography
[1] W. Bludau et al. J. Appl. Phys., (45):1846, 1974.
[2] Centrovision, A Primer on Photodiode Technology, 2000.
URL http://www.centrovision.com/tech2.htm.
[3] H. Graf. Intelligent Cameras - The future is here. Image Processing Europe,
28 32, Oktober 1997.
[4] R. G. Humphreys et al. Phys. Rev. B., (18):5590, 1978.
[5] D. Jansen. Optoelektronik. Vieweg Verlag, 1993. ISBN 3-528-04714-3.
[6] H. Keller, M. Boehm. TFA (Thin Film on ASIC) Image Sensors. In: R. J. Ahlers
(Hrsg.), 6. Symposium Bildverarbeitung 99, 41 49. Technische Akademie Esslin-
gen, Technische Akademie Esslingen, November 1999. ISBN 3-924813-43-4.
[7] J. Lienhard. 3D-Scanner, Entwicklung eines Verfahrens zur dreidimensionalen
Objekterfassung. Master s thesis, University of Applied Sciences, Offenburg, Ger-
many, 1995.
[8] F. Lukes, E. Schmidt. In: Int. Conf. Phys. Semicond. Exeter, 1962,, number 45,
389, London, 1962. Inst. of Physics, London.
[9] G. G. Macfarlane et al. Phys. Rev., (108):1377, 1957.
[10] M. Marso. Poröses Silizium. Technical Report 12, Forschungszentrum Jülich
GmbH, Technologie-Transfer-Büro, November 1998. 189-206 pp.
[11] K.. Meier. CMOS Photorezeptoren und On-Chip Bildverarbeitung. In: B. Jähne
(Hrsg.), 2. Heidelberger Bildverarbeitungs-Forum: Sensorik, 55 66, IWR, Uni-
versität Heidelberg, Im Neuenheimer Feld 368, 69120 Heidelberg, März 1996.
Forschungsgruppe Bildverarbeitung, Jähne, B.
[12] A. Moini, Vision chips or seeing silicon, 1997.
URL http://www.eleceng.adelaide.edu.au/Groups/GAAS/Bugeye/visionchips/.
[13] M. Preier. Appl. Phys., (20):189, 1979.
[14] R. B. Schoolar, J. R. Dixon. Phys. Rev., (137):667, 1965.
[15] P. Seitz. Stand und Perspektiven der Halbleitersensorik. In: B. Jähne (Hrsg.), 2.
Heidelberger Bildverarbeitungs-Forum: Sensorik, 11 16, IWR, Universität Hei-
delberg, Im Neuenheimer Feld 368, 69120 Heidelberg, März 1996. Forschungs-
gruppe Bildverarbeitung, Jähne, B.
Copyright 2000 A. Erhardt-Ferron
52 Bibliography
[16] D. D. Sell et al. J. Appl. Phys., (45):2650, 1974.
[17] V. V. Sobolev et al. Sov. Phys. Semicond., (12):646, 1978.
[18] H. Stiebig. Mehr-Kanal-Sensoren zur Farberkennung. Technical Report 43,
Forschungszentrum Jülich GmbH, Technologie-Transfer-Büro, 1999.
[19] M. Wülker, Physik, 1999.
URL http://mv-sonne.m.fh-offenburg.de/Physik/.
[20] CBV. Basiskatalog. Firmenkatalog, Computer BV, Bildverarbeitung GmbH, Lud-
wigshafener Str. 4b, D-67227 Frankenthal.
[21] DBS. The Imaging Source. Firmenkatalog, DBS Digitale Bildverarbeitung und
Systementwicklung GMBH, Kohlhökerstr. 61, 28203 Bremen, Tel. [phone]-
0;, Frühjahr 1998.
[22] DBS. Digitale Bildverarbeitung. Firmenkatalog, DBS Digitale Bildverarbeitung
und Systementwicklung GmbH, Kohlhökerstr. 61, 28203 Bremen, Tel. 0421-
33591-0, 1999. Katalog bei http://www.dbs.de/ in English.
[23] PCO, Know How, 1999.
URL http://www.pco.de/.
Copyright 2000 A. Erhardt-Ferron